mlsanityData to 7 Oct 2026

Forget Unlearning: Towards True Data-Deletion in Machine Learning

arXiv 2210.08911 ↗First posted 17 Oct 2022stat.ML cs.LGcited by 58

On the map: LLM safety and security → Machine unlearning

Abstract (arXiv)

Unlearning algorithms aim to remove deleted data's influence from trained models at a cost lower than full retraining. However, prior guarantees of unlearning in literature are flawed and don't protect the privacy of deleted records. We show that when users delete their data as a function of published models, records in a database become interdependent. So, even retraining a fresh model after deletion of a record doesn't ensure its privacy. Secondly, unlearning algorithms that cache partial computations to speed up the processing can leak deleted information over a series of releases, violating the privacy of deleted records in the long run. To address these, we propose a sound deletion guarantee and show that the privacy of existing records is necessary for the privacy of deleted records. Under this notion, we propose an accurate, computationally efficient, and secure machine unlearning algorithm based on noisy gradient descent.

Key facts. Cited by 58 papers in the map (data through 7 Oct 2026). Placed in Machine unlearning, one of 12 directions in LLM safety and security. Builds on 15 of its 43 references that are in the map. First posted on arXiv 17 Oct 2022.

Builds on

15 of 43 references.

Closest ancestors · 3

Same direction, newest first

  1. 2021Adaptive Machine Unlearning8 Jun
  2. 2021Remember What You Want to Forget: Algorithms for Machine Unlearning4 Mar
  3. 2021Machine Unlearning via Algorithmic Stability25 Feb

Other references · 12

  1. 2016Deep Learning with Differential Privacycited by 2,682
  2. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,221
  3. 2019Machine Unlearningcited by 1,020
  4. 2017Renyi Differential Privacycited by 643
  5. 2019Certified Data Removal from Machine Learning Modelscited by 512
  6. 2019Making AI Forget You: Data Deletion in Machine Learningcited by 447
  7. 2020Descent-to-Delete: Gradient-Based Methods for Machine Unlearningcited by 303
  8. 2020Approximate Data Deletion from Machine Learning Modelscited by 250
  9. 2019Rapid Convergence of the Unadjusted Langevin Algorithm: Isoperimetry Sufficescited by 224
  10. 2020When Machine Unlearning Jeopardizes Privacycited by 181
  11. 2021Differential Privacy Dynamics of Langevin Diffusion and Noisy Gradient Descentcited by 71
  12. 2015Privacy for Free: Posterior Sampling and Stochastic Gradient Monte Carlocited by 61

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 1; 2023: 7; 2024: 19; 2025: 18; 2026: 13.

58 papers.

Sort:

By area:

  1. 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Modelscited by 251
  2. 2024Machine Unlearning of Pre-trained Large Language Modelscited by 118
  3. 2024Langevin Unlearning: A New Perspective of Noisy Gradient Descent for Machine Unlearningcited by 48
  4. 2024Machine Unlearning for Image-to-Image Generative Modelscited by 47
  5. 2023Exploring the Landscape of Machine Unlearning: A Comprehensive Survey and Taxonomycited by 44
  6. 2022Hidden Poison: Machine Unlearning Enables Camouflaged Poisoning Attackscited by 37
  7. 2024Machine Unlearning Fails to Remove Data Poisoning Attackscited by 35
  8. 2024Threats, Attacks, and Defenses in Machine Unlearning: A Surveycited by 34
  9. 2023SAFE: Machine Unlearning With Shard Graphscited by 28
  10. 2024Separable Multi-Concept Erasure from Diffusion Modelscited by 25
  11. 2025Certified Unlearning for Neural Networkscited by 25
  12. 2024Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerablecited by 24
38 more
  1. 2024The Utility and Complexity of in- and out-of-Distribution Machine Unlearningcited by 20
  2. 2024Learning to Unlearn for Robust Machine Unlearningcited by 19
  3. 2024DiffZOO: A Purely Query-Based Black-Box Attack for Red-teaming Text-to-Image Generative Model via Zeroth Order Optimizationcited by 19
  4. 2023Training Data Protection with Compositional Diffusion Modelscited by 16
  5. 2024Certified Machine Unlearning via Noisy Stochastic Gradient Descentcited by 13
  6. 2023Adapt then Unlearn: Exploring Parameter Space Semantics for Unlearning in Generative Adversarial Networkscited by 11
  7. 2025Towards Lifecycle Unlearning Commitment Management: Measuring Sample-level Unlearning Completenesscited by 11
  8. 2025Distillation Robustifies Unlearningcited by 11
  9. 2023Ticketed Learning-Unlearning Schemescited by 10
  10. 2025When to Forget? Complexity Trade-offs in Machine Unlearningcited by 9
  11. 2025Rectifying Privacy and Efficacy Measurements in Machine Unlearning: A New Inference Attack Perspectivecited by 8
  12. 2024Towards Lifecycle Unlearning Commitment Management: Measuring Sample-level Approximate Unlearning Completenesscited by 6
  13. 2025MUBox: A Critical Evaluation Framework of Deep Machine Unlearningcited by 6
  14. 2023Fast-NTK: Parameter-Efficient Unlearning for Large-Scale Modelscited by 5
  15. 2025Erasing Without Remembering: Implicit Knowledge Forgetting in Large Language Modelscited by 5
  16. 2025Distributional Machine Unlearning via Selective Data Removalcited by 5
  17. 2024A Reliable Cryptographic Framework for Empirical Machine Unlearning Evaluationcited by 4
  18. 2025Machine Unlearning under Overparameterizationcited by 4
  19. 2026The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samplescited by 4
  20. 2024Partially Blinded Unlearning: Class Unlearning for Deep Networks a Bayesian Perspectivecited by 3
  21. 2024Goldfish: An Efficient Federated Unlearning Frameworkcited by 3
  22. 2025System-Aware Unlearning Algorithms: Use Lesser, Forget Fastercited by 3
  23. 2026Class Unlearning via Depth-Aware Removal of Forget-Specific Directionscited by 3
  24. 2023Towards Regulatable AI Systems: Technical Gaps and Policy Opportunitiescited by 2
  25. 2024Erase to Enhance: Data-Efficient Machine Unlearning in MRI Reconstructioncited by 2
  26. 2024Data Deletion for Linear Regression with Noisy SGDcited by 2
  27. 2025TAIJI: MCP-based Multi-Modal Data Analytics on Data Lakescited by 1
  28. 2025Apollo: A Posteriori Label-Only Membership Inference Attack Towards Machine Unlearningcited by 1
  29. 2025Efficient Unlearning with Privacy Guaranteescited by 1
  30. 2025One-Point Contraction: Erasing Representational Separability toward Irreversible Deep Forgettingcited by 1
  31. 2025Forget What's Sensitive, Remember What Matters: Token-Level Differential Privacy in Memory Sculpting for Continual Learningcited by 1
  32. 2025Fully Decentralized Certified Unlearningcited by 1
  33. 2026Protecting the Undeleted in Machine Unlearningcited by 1
  34. 2026Less Noise, Same Certificate: Retain Sensitivity for Unlearningcited by 1
  35. 2026Jellyfish: Zero-Shot Federated Unlearning Scheme with Knowledge Disentanglementcited by 1
  36. 2026Auditing of Unlearning Algorithmscited by 1
  37. 2024Game-Theoretic Machine Unlearning: Mitigating Extra Privacy Leakagecited by 0
  38. 2025Machine Unlearning for Streaming Forgettingcited by 0

50 most cited of 58 All 58 (JSON)